Dissatisfaction with IP multicast's slow start seems to be growing, especially within the commercial sector. Recently the IETF chartered an ad hoc group, called maddogs, to explore ways to overcome the problems of IP multicast and to make it more suitable for commercial products and services.
In spite of this, however, many people seem still to accept the assumption that IP multicast should be the foundation for internet distribution. In support of this statement is the fact that, in addition to the substantial inter-domain multicast IP protocol work going on, there are research and working groups for reliable multicast transport.
There are a couple of exceptions. Researchers at Telcordia and the University of Maryland, based on a suggestion by Christian Huitema, have independently proposed the idea of host-based multicast. They developed a protocol for use in the ad hoc mobile network environment. The protocol is called AMRoute, and has been published as an internet draft in the Manet working group of IETF. This specific document targets IP multicast functionality only (by which I mean it is works across space only), and for mobile ad hoc networks only. The internet draft does, however, mention in passing that the approach may have broader functionality in the larger internet. As far as I know, however, no substantial work in this direction has been done.
Hui Zhang at Carnegie Mellon has also independently been working on host-based multicast (which he calls endsystem-only multicast). He and his students have worked out some basic low-level tree-formation mechanisms and have done some simulations (see Section 2.4). They recognize most of the broader issues, such as naming, and have independently conceived many of the ideas in yoid (such as the rendezvous), though they haven't done much work on these broader issues per se.
More recently, the Reliable Multicast Research Group (RMRG) of the IRTF has started a work item on what it so far calls Auto Tree Configuration. Among the goals of that effort is to create an algorithm that can build a tree strictly among the recipient hosts of a multicast group. As near as I can tell, however, the purpose of that tree is not to distribute content per se, which is still expected to be transmitted via IP multicast. Rather, the tree is for transmitting acks/naks and optionally for local retransmissions of packets dropped by IP multicast.